Divergent Dynamics of Epigenetic and Genetic Heterogeneity in Relapsed Acute Myeloid Leukemia

Acute Myeloid Leukemia (AML) remains a clinical challenge, with most patients dying of relapsed disease. The complete biological basis of relapse remains unclear.  Genetic lesions and heterogeneity have been proposed as key drivers of clinical outcome, yet do not fully explain leukemia relapse.  Epigenomic dysregulation is a hallmark of newly diagnosed AML.  Plasticity is a core property of the epigenome, enabling cells to adapt to stressful conditions, independent of genetic alterations. Hence we asked whether epigenomic plasticity might contribute to AML progression, have functional consequences and be independent of genetic influences in AML (a question that has not been addressed for any tumor type).

Methods. We formed an international consortium to collect and profile paired diagnosis and relapse AML specimens. We extracted DNA and RNA from 138 clinically annotated AML patient samples.  We obtained matched germline DNA as genetic controls, and fourteen normal CD34+ specimens as DNA methylation and transcriptome controls. We performed methylome sequencing (ERRBS), genomic sequencing (exomes and targeted resequencing) and transcriptomic (RNA-seq) profiling. For a single patient, more intensive multi-layer profiling (whole genome sequencing, ERRBS, RNA-seq and single cell RNA-seq) was performed at five serial time points.  We quantified epigenetic allelic heterogeneity (epialleles) using a novel approach that employs entropy equations (MethClone), and validated epiallele composition using orthogonal methods.

Some of the major conclusions are:

1) Epigenetic allelic diversity is an independent variable linked to clinical outcome.Statistically significant epiallele shift (ΔS <-90) was detected at thousands of genomic loci (eloci) at diagnosis.  High eloci burden correlated (Wilcoxon test) with a shorter relapse free probability in the entire cohort  (p = 0.043) and in intermediate-risk patients based on the Medical Research Council (p= 0.016) and European Leukemia Net (p=0.057) criteria. Multivariate analysis using Cox proportional hazards regression model revealed that the epiallele burden was an independent variable correlated with relapse free survival (p = 0.021).

2) Promoter epialleles are linked to hypervariable transcriptional regulation.   We observed substantial change in epiallele burden at relapse versus diagnosis. A subset of the eloci localized to gene promoters. High promoter epiallele variance was significantly associated with high transcriptional variance (p<0.001) based on RNA-seq, including genes that were significantly differentially expressed at relapse.  Deconvolution of leukemia blast populations using Single Cell RNA-seq confirmed that the presence of promoter epialleles was linked to hypervariable transcriptional states (p<0.001).

3) AML patients can be classified according to epigenetic allele progression at relapse.  K-means clustering based on epiallele shift at diagnosis versus relapse distributed patients into three classes:  those with reduced, increasing or stable epiallele burden.  Strikingly, there was no correlation between epiallele changes and the patterns of genomic evolution. Furthermore, there was no correlation between epiallele patterns acquired with mutations in epigenetic modifiers or other recurrently mutated genes in AML.

4) Epigenetic heterogeneity upon disease relapse is divergent from the genetic landscape. Integrating whole genome sequencing and methylome analysis we observed that a) significant increases in epigenetic heterogeneity precede significant changes in the abundance of somatic mutations; b) whereas a high number of somatic mutations were shared across all time points, epialleles exhibited dominance of distinct and unique eloci at each time point; and c) the variant epiallele frequency decreased earlier in progression than somatic mutation variant allele frequency, suggesting that epigenetic clonal diversification can precede genetic clonal evolution.  

Summary. Based on our results we propose that epigenetic allele diversity allows populations of leukemia cells to sample transcriptional states more freely thus creating the potential for greater evolutionary fitness.  This provides an additional independent mechanism of plasticity that can explain the resilient nature of AML to adapt and survive exposure to chemotherapy drugs, independent of genetic heterogeneity.